Professional Documents
Culture Documents
Yang 2015
Yang 2015
fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
Abstract—RFID technology has been widely used for object tracking in indoor environment due to their low cost and convenience
for deployment. In this paper we consider RFID reader tracking which refers to continuously locating a mobile object by attaching
it with a RFID reader that communicates with passive RFID tags deployed in the environment. One difficulty is that the RFID
readings gathered from the environment are often noisy. Existing approaches for tracking with noisy RFID readings are mostly
based on using Particle Filter (PF). However, continuous execution of PF has extremely high computational cost, and may be
difficult to be done on mostly resource constrained mobile RFID devices. In this paper, we propose a hybrid method which
combines PF with Weighted Centroid Localization (WCL) to achieve high accuracy and low computational cost. Our observation
is that WCL has the same accuracy with PF with much lower cost if the object’s velocity is low. Our method has two critical
features. The first feature is adaptive switching between using WCL and PF based on the estimated velocity of the mobile object.
The second feature is the further reduction of computational cost by offloading costly PF algorithm onto nearby servers. We
evaluate the performance of our method through extensive simulations and experiments in two real world applications, namely,
indoor wheelchair navigation and in-station Light Rail Vehicle (LRV) tracking at one of Hong Kong MTR depots. The result shows
that our proposed approach has significantly less computational cost than existing PF based methods, while being as accurate
as them.
1 I NTRODUCTION reader
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
Second, the RFID readings gathered from real world pattern of the moving object (e.g. orientation) becomes
are noisy. It means that each tag in the reading range stable under high velocity. This stable pattern could
of the RFID reader is not certainly but possibly to be be taken advantage of to improve the sampling ef-
detected. Detection failure of tags is normal. The de- ficiency, if particle filter is used, and thus to reduce
tection failures happen frequently and normally, while the computational cost. The complementary property
in other localization systems, e.g., infrared based sys- of PF and WCL in terms of accuracy and efficiency
tem, ultrasonic based systems and WiFi based sys- motivates us to integrate them together.
tems, the loss of signals from sensors/beacons did Second, our hybrid method can offload
not happen normally. The challenge is to develop intensive computations required by PF onto
new approaches that can tolerate high uncertainty nearby infrastructures or servers based on the
of the RFID readings. Third, to tolerate the detec- quality of network connections. We evaluate our
tion failures in passive RFID, an effective approach solution through real world experiments, and show
is to repeat the scanning periodically with a high that our proposed approach with computation
frequency. Redundant high-speed scanning of tags offloading outperforms other approaches in terms
can improve the accuracy under uncertain sensory of both accuracy and efficiency. We summarize the
readings, but it causes high computational cost, which contribution of this paper as follows:
may exceed the processing capability of the resource
constrained mobile devices. Thus, the challenge is to • We propose a hybrid method for object tracking
design low computational cost methods that could be using noisy passive RFID readings. Compared
accommodated by the mobile RFID devices. with existing particle filter method, it saves a lot
Existing methods for tracking mobile objects using of computational cost while achieving the same
noisy RFID readings are often based on Kalman filter accuracy.
[19], and Particle Filter (PF) [3][6][13][14], also known • We design a fast and cheap method to estimate
as Sequential Monte Carlo Method [20]. Kalman filter mobile object’s velocity from the RFID readings.
is suitable to be used in linear system with Gaus- It is not necessarily applied in our proposed hy-
sian noise. However, in many practical RFID tracking brid method for object tracking, but can be used
systems, the location evolves none-linearly over time, separately in other application scenarios which
and the noise may not be Gaussian. In these none- require the measurement of object’s speed.
linear none Gaussian systems, particle filter can get • Besides extensive simulations, we have validat-
more accurate solution than Kalman filter. In particle ed the method in two real systems, indoor
filter, the object’s position is represented by a set of wheelchair tracking and LRVs (Light Rail Vehi-
weighted samples (called particles). Two procedures, cles) tracking in one of Hong Kong MTR depots.
sampling and filtering, are executed repeatedly. Dur-
ing the sampling, the particles are sampled randomly
2 M ODELS
from an area which is usually predicted according
to the motion model of the mobile object. During 2.1 System Model
the filtering, each particle is first assigned a weight The RFID reader based tracking system contains three
according to the sensing model. The particles with component: RFID reader, a number of passive RFID
low weight are then filtered out. However, continuous tags, and the environment infrastructures, which are
execution of particle filter suffers from high compu- shown in Fig.1(b). The reader is attached on the
tational cost on a resource constrained RFID enabled object to be tracked. The infrastructures include the
device. Other existing approaches [1] [2] [10] such as servers deployed in the area where the object is
Weighted Centroid Localization (WCL) are cheap in moving. The RFID reader can communicate with the
computational cost, but yield bad accuracy specially server through wireless networks such as WLAN.
when the object’s speed is high. We assume that the RFID reader is integrated with
In this paper, we propose a hybrid method for a circularly polarized antenna. Fig.2 illustrates the
tracking mobile objects with high accuracy and low geometrical relationship of the reader and tags. The
computational cost. This method particularly uses two reader has a height of H above the ground, and
critical techniques. First, it adaptively chooses costly a circular reading range tr. The number of passive
PF and cheap WCL according to the estimated speed tags deployed in the tracking area are K. Each tag’s
of the object. If the mobile object’s velocity is less location, denoted as Ti , is stored in advance at the
than a threshold, WCL is used; otherwise, the particle reader in association with its tag ID. Whenever the
filter is used. This is because of our observation that, reader detects a tag, it can access the tag’s location
when the object’s speed is low, WCL is able to achieve through the tag ID. The target moves with a velocity
at least the same accuracy with particle filter while of v. The RFID readings are obtained in a frequency
costing much less computation. Oppositely, when the of f .
object’s speed is high, the accuracy of WCL is much It is assumed that detection failure could happen
worse than particle filter. Furthermore, the motion even when the tag appears in the reader’s reading
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
0.15
where λt (k) is temporal weight of RFID reading at
0.1
time k. In our method, temporal weights are deter-
0.05
mined by intersection of reading range at time k and
0
that of time t. 0.2 0.4 0.6 0.8 1 1.2
Tag density (divided by tr)
1.4
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
square pattern or triangular patter, does not affect the • Updating Particles are reweighted according to
accuracy. the latest RFID reading zk and the measurement
model, Equation(11-12).
3.3 Particle Filtering wki = wk−1
i
p(zk |lki )
In this subsection, we solve this object tracking prob-
The weight is then normalized by wki =
lem through a Bayesian approach [8]. Using this PM
approach, two models are generally required: first, a wki / i=1 wki .
model describing the evolution of the object’s location • Resampling A new set of n particles with e-
with time (the system model); second, a model relat- qual weights 1/n is obtained from the old set of
ing the noisy RFID readings (the measurement mod- particles. The possibility of choosing particle i is
el). These models are presented in a probabilistic form. proportional to its weight wki . The objective of
Considering the evolution of the position sequence of resampling is to reduce the effects of degeneracy.
a moving object, lk = (xk , yk , θk ), k = 1, 2, 3, · · ·, we The basic idea is to eliminate the particles that
have the following system model have small weights and to concentrate on parti-
cles with large weights.
v
xk = xk−1 + cos θk−1 (8)
f
3.4 Adaptive Algorithm Selection of WCL and PF
v
yk = yk−1 + sin θk−1 (9) 3.4.1 Limitation of WCL
f
WCL faces two weaknesses in object tracking. First, if
θk = θk−1 + U (−∆θ, ∆θ) (10)
the tags are deployed sparsely onto the ground, it is
where xk , yk are two coordinates of the object’s possible that tr can not achieve the necessary value
location, θk is the heading orientation of the objec- 1.1a (see the discussion above about Fig.5), even when
t, U (−∆θ, ∆θ) is an increase angle of the heading the Reader’s power reaches the maximum level. In
direction which yields a uniform distribution, v is this case, the accuracy of WCL will be degenerated
the velocity of the object, and f is the observation seriously, as shown by Fig.5. Second, the observation
frequency of the reader. The measurement model is frequency of RFID f always has a up-bound in order
given by not to affect the detection rate of tags. Given this
M
Y upper bound of f , if the object moves in a high
p(zk |lk ) = p(rki |lk ) (11) speed, most of the measurements in the past become
i=1 invalidate to calculate the current location. Therefore,
p(rki |lk ) = rki p(di ) + (1 − rki )(1 − p(di )), (12) the accuracy will decrease as the speed of the object
increases, which is also demonstrated by Fig.7 in
where di is the distance of tag i from target and p(di ) Section 4.
is the RFID sensor model described in Section 2. The
objective is to recursively estimate lk from the RFID 3.4.2 Limitation of Particle Filter
readings, denoted as zk = {rki }Mi=1 . Although Particle Filter (PF) has been widely used
From a Bayesian perspective, the tracking problem in object tracking for its satisfactory accuracy, the
is to recursively calculate some degree of belief in the biggest limitation of PF comes from its high compu-
state lk at time k, given the data z1:k up to time k. It tational cost. Table 3 in Section 4 demonstrates its
is required to construct the posterior pdf , usually in costliness in computation. Since mobile objects are
two stages: prediction and update. usually equipped with resource constrained devices,
A particle filter approximates the posterior pdf at continuously running PF on these devices may not be
time k by a set of M particles. Each particle contains possible.
a position hypothesis lki = (xik , yki , θki ) and a weight
wki . wki indicates the importance of the i-th particle. A 3.4.3 The hybrid method to integrate WCL with PF
position estimate lˆk can be obtained by lˆk = i=1 wki lki
PM
Aware of limitations of both WCL and PF, the ques-
. The particle filtering algorithm takes the following tion here is, can we combine these two methods
steps in a recursive fashion: together in order to achieve both high accuracy and
• Prediction The target position at time k is predi- efficiency in object tracking? To realize this objective,
cated by propagating all particle positions at time our strategy is that in any case WCL does not have
k − 1 according to system model, Equation(8-10). worse accuracy than PF, WCL will be adopted to
replace PF in object tracking. These cases are formu-
lki ∼ p(lk |lk−1
i
, v, ∆θ)
lated by the following conditions: trmax >= 1.1a and
We assume that v has a Gaussian distribution, v <= vth = 2tr·f
Nth , where vth is the velocity threshold,
N (vk , σk2 ). The model probabilistically describes Nth is the required duration (denoted as the number
the transition of system from old state for a given of time slots) by WCL that a tag can stay in the reading
measurement vk . range of a moving object.
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
...
the hybrid method. If Nth is set too high, that means
the threshold of the speed vth is too low, the hybrid RFID Device Server
method has little improvement on the computational
cost compared with particle filter. Oppositely, if Nth is Fig. 6. The data transmission between RFID device
set too low, the hybrid method sacrifices the accuracy and server if particle filter is executed remotely
too much. It is hard to prove theoretically the optimal
value of Nth . However, we can acquire an optimal Section 5. The main processing platform of the RFID
value of Nth from the numerical simulations. Section device in our implementation is the Mica2 sensor
4 demonstrates an optimal value of Nth in a simulated mote. The remote server is the Fujitsu LifeBook S
scenario. Series. Table 2 shows the one time computational cost
of WCL and PF respectively in location and remote
3.5 Computation Offloading execution. The execution of WCL at both the mobile
To further reduce the computation complexity on device and the server causes very little computation
the RFID device, our method can offload the cost. For simplicity in our system, we only consider
computation-intensive part of the tracking algorithm the offloading of PF. Note that the execution time of
to a nearby server. Although offloading computation PF varies depending on the number of particles. In
onto the server can lower the computational cost on Table 2, the execution time of PF is measured when
the RFID device, it brings additional communication the number of particles is 2000.
cost due to the transmission of required data between To make the optimal offloading decision, we need
the RFID device and the server. The principle of the to estimate the communication cost of PF when it is
offloading decision is that, given the algorithm func- executed remotely. At each time of location estima-
tion, if reduced computation cost by remote execution tion, the PF needs to predict the state of the particles
is large than the incurred communication cost, then according to the mobility model of the object, and
it is better to offload this function onto the server; then update the weights of the particles based on the
otherwise the function is better to be executed locally. RFID readings in this time slot. Fig.6 shows the data
•Offloading Principle. Assume the estimation of transmission between the RFID device and the server
location at time t with the RFID readings zt needs when PF is performed remotely. At the beginning of
time τl if it is performed locally on the RFID device, the time slot t, the RFID device uploads the RFID
and needs time τr if it is executed remotely on the readings zt to the server. The server does the predic-
server. Let ds denote the size of transmission data tion and resampling, and returns the state of particles
in remote execution. Let B denote the bandwidth of as well as the estimated location at time t. The data
the wireless connection between the RFID device and required to transfer on the network mainly includes
the server. We have the following offloading decision. the observed RFID readings zt , the state of particles,
If τm − τr > dBs , the algorithm is then offloaded to i.e., the parameters wki and lki . The reason that the
the server at this time k; otherwise, the algorithm is state of particles need to be returned to the RFID
executed on the RFID device. In the following, we device is that the wireless network connection may be
describe how to collect τr , τl , ds and B in real systems. intermittent. We need to backup the state on the RFID
In the hybrid method, the WCL and PF algorithm- device. In case the network connection is down, the
s are adaptively executed according to the object’s PF can be continued locally with the backup particles
velocity. We collect both the local and remote com- state. Note that the size of transmission data increases
putation cost of WCL and PF from our experimen- as the number of particles in PF increases.
tal implementation, which is described in details in With the size of transmission data, we need to mea-
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
0.2
TABLE 3
PF (Particle number = 2000) Efficiency Comparison between WCL and PF
0.18 PF (Particle number = 1000)
PF (Particle number = 500)
0.16
WCL v(m/s) 0.1 0.2 0.3 0.4 0.5 0.6
Location Error (m)
0.14
WCL(ms) 119 86 75 64 65 66
0.12
PF-2000(ms) 7593 8484 10568 9792 9260 9276
0.1
PF-1000(ms) 4654 4927 6012 4873 4825 5103
0.08
0.06
PF-500(ms) 2769 2876 3421 2785 2631 2724
0.04
0 0.1 0.2 0.3 0.4 0.5 0.6
Velocity (m/s)
Fig.7 compares the location error of WCL and PF
Fig. 7. Accuracy comparison between WCL and PF under different velocities. We plot the location error
of PF when the number of particles are respectively
sure the bandwidth of the wireless network, in order 2000, 1000, and 500. We can see that the location error
to estimate the communication cost. In our method, of WCL increases as the velocity increases while the
we periodically estimate the network bandwidth by accuracy of particle filter has no association with the
transferring a 10K file between the device and server. velocity. PF has better accuracy than WCL when the
The bandwidth is the testing file size divided by the velocity exceeds 0.2m/s. When the velocity is less
transmission time. In our experiment, the RFID con- than the value of 0.2m/s, WCL has almost the same
nects the server through ZigBee motes. We measure accuracy with PF. Fig.7 also shows that the accuracy of
the bandwidth fluctuation through carrying with the PF becomes better as the number of particles increas-
RFID device and walking around in one classroom. es. Table 3 compares the execution time of the two
We found that the bandwidth indeed fluctuates. The methods. Simulations for both algorithms are run on
fluctuation period is on the order of several seconds. the same platform (Matlab Environment on a Fujitsu
LifeBook S Series). To compare the computational
4 P ERFORMANCE E VALUATION THROUGH efficiency among the cases of different velocity, the
number of trajectories are tuned for each velocity so
S IMULATIONS that the tracking time in all the cases keeps the same.
In this section we evaluate the performance of the From Table 3, we can see that WCL is obviously more
proposed hybrid method. We consider the following efficient than PF whatever the target’s velocity is. The
metrics: location error (accuracy) and computational PF has more computational cost as the number of
cost (efficiency). First, we evaluate respectively and particles increases.
compare the performance of WCL and Particle Filter Fig.8a and Fig.8b show the location error at each
(PF). The parameter we consider in the comparison is time slot as well as the estimated trajectory when the
velocity because it is the main factor that distinguishes object’s velocity is 0.6m/s. We can see that the location
the performance of these two methods. Then, we eval- error of WCL is larger than that of PF. It is interesting
uate how much the hybrid method can improve the to note that the location deviation of WCL is along
performance by considering a trajectory along which the trajectory while the location deviation of PF is
the object moves with a varying speed. We further perpendicular to the trajectory. This is because that
evaluate the impact of the computation offloading executing the WCL periodically for object tracking has
strategy onto the performance. a critical limitation. If no tag is detected in current
time slot, the tags detected in the past time slot
will dominate the centroid estimation of the current
4.1 Performance Comparison between WCL and position. Therefore, the result at current time is very
Particle Filter likely to be the real location in the past time slot. In
The simulated environment is a 4m × 4m plane. The specific, this location error becomes very serious when
RFID tags are densely deployed onto the plane. The the object’s speed is high.
spacing size of tags a is 0.27m. The reading range
of the reader tr is 0.3m. The observation frequency
f is 10Hz. Assume that the object moves with a 4.2 The Hybrid Method of WCL and PF
constant speed around a rectangle trajectory in the We compare the performance of the hybrid method
2-Dimension plane. The RFID model described in with that of other two methods, WCL and Particle
Section 2.2 is used to generate the simulated RFID Filter, by considering a more practical scenario where
readings. The location is estimated continuously dur- the object’s velocity varies during the trajectory. Fig.8c
ing one trajectory through different tracking methods. shows a simple case that the object circulates around
The accuracy is represented by an average of the the rectangle with two different velocities alternative-
location error at each time slot. ly. Around four corners the object moves with a slow
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
Fig. 8. (a)(b) visualize the location error at each time slot as well as the entire estimated trajectory when the
object’s velocity keeps constant at v = 0.6m/s; (c) is a simulated trajectory with a varying velocity
TABLE 4 200
Bandwidth (KB/s)
High power
Performance comparison between the three methods 150 Low power
0
ExecutionTime(s) 0.08 4.53 1.26 0 20 40 60 80 100
Time (s)
velocity, v = 0.15m/s. In other intervals the object Fig. 9. Network bandwidth trace
runs with a fast velocity, v = 0.6m/s.
How to select the threshold vth and Nth when the •Hybrid-Local. In this method, the tracking algo-
hybrid method is performed in the scenario? It has rithm is adaptively selected from WCL and PF. The
been shown in Fig.7 that the optimal vth is 0.2m/s. algorithm can be offloaded onto the server.
Below this threshold, WCL has the same accuracy In this evaluation, the object’s moving trajectory
with PF, but above it WCL has much worse accuracy and velocity have the same setting with Fig.8c. The
8 8
than PF. The Nth is computed by Nth = 2tr·f time used in this trajectory is 0.15 + 0.6 = 66.7s. As the
vth = 30.
Table 4 presents the location error and execution sampling frequency is 10 Hz, we have about 667 time
time of the three methods. Note that the number of slots in the whole trajectory. The number of particles
particles is set as 2000 in all PF executions. It shows in all PF executions is set as 2000. Table 2 shows the
that hybrid method leads to much lower location error profiling computational cost of WCL and PF in each
than WCL. Moreover, compared with particle filter, step (or time slot) respectively on the RFID device and
the hybrid method saves about 75% computational the server. Since WCL runs very fast on local device,
cost while achieving the same accuracy. we do not consider the offloading of WCL. When PF
is offloaded onto the server, we measure that the size
of transmitted data between the RFID device and the
4.3 The Offloading Strategy service is 5.2 kB.
In this section, we evaluate the influence of offloading We collect real network bandwidth traces from one
strategy onto performance, particularly in term of classroom. In the classroom, one notebook is deployed
the efficiency. We compare the performance of the as the server. The notebook is connected with a TelosB
following five tracking methods. mote. The RFID device attached with Mica2 mote
•WCL-Local. In this method, the tracking algorith- communicates with the server through ZigBee. We
m is WCL. The algorithm is executed on the RFID carry the device and move randomly around the
device. classroom. The network bandwidth is measured by
•PF-Local. In this method, the tracking algorithm is transferring one 10kB testing file. The measurement
Particle Filter (PF). The algorithm is executed on the is done every 5 seconds. Fig.9 shows two network
RFID device. traces when the transmission power of the ZigBee
•PF-Offloading. In this method, the tracking algo- motes is set with various levels. In high power setting,
rithm is Particle Filter (PF). The execution of PF can both motes have level 256 in transmission power,
be offloaded onto the server at each time slot. while in low power setting, both motes have level 128
•Hybrid-Local. In this method, the tracking al- in transmission power. We use the network trace in
gorithm is adaptively selected from WCL and PF high power setting to evaluate the performance with
according to the estimated velocity. The algorithm is offloading strategy. Note that only the first 66.7s of
executed on the RFID device. the whole network trace is used in the evaluation.
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
250
250
200
Efficiency (ms)
50
0.4
0.3 0
0 50 100 150 200 250 300
0.2 Wireless bandwidth (KB/s)
0.1
3 4
Hybrid-Offloading has the best performance among
5 7
the five methods. 6
10
We then evaluate how the computational efficiency
of the methods changes depending on the band- 2
width of wireless connection between the RFID de-
vice and the server. Fig.11 plots the efficiency as the
bandwidth increases from zero to 300kB. The two
methods with offloading strategy, Hybrid-Offloading Fig. 13. Wheelchair trajectories in a 4m × 6m class-
and PF-Offloading, have decreasing execution cost as room
the bandwidth increases. This is because under the
offloading strategy, large bandwidth leads to short the tags deployment. The reader contains 18 discrete
time in the data transmission between the device power levels. We tune the power on level 12 so that
and server. When the bandwidth is as little as zero, the reading range tr is as far as 60cm, satisfying the
the methods with offloading strategy have the same equation tr >= 1.1a, where a = 50cm.
execution cost with the methods running locally on The wheelchair can be controlled with a joystick
the device. controller. We attach an extension unit to the joystick
for controlling the output signal. This unit is connect-
5 E XPERIMENTS ed to a Mica2 Mote through serial port. The speed of
the wheelchair is controlled by a pair of output volt-
5.1 Indoor wheelchair tracking age values. These voltage values can be set through
In the scenario of wheelchair tracking, the RFID tags the Mica2 Mote. In our experiment, the wheelchair
are densely deployed on the ground in a 4m × 6m moves along a line with a varying speed. During
classroom. The spacing size of tags (density) is 50cm. the first half of the entire trajectory, the wheelchair’s
The RFID device we attached on the chassis of the speed is set to 0.1m/s. In the second half, the speed
wheelchair is an integrated UHF RFID reader with a is increased to 0.6m/s. The observation frequency f
circularly polarized antenna. Fig.12 and Fig.13 shows is 10Hz. By selecting Nth = 40, the threshold of
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
TABLE 5
Experiment results for wheelchair tracking
Methods WCL PF-O Hybrid-O Hybrid-L
Location error (m) 0.482 0.186 0.195 0.193
Time per step (ms) 3.6 89.2 19.4 95.7
0.9
(a) (b)
0.8
0.7
Fig. 15. The deployment of RFID system at one MTR
depot
0.6
CDF (%)
0.5
0.4
[20] along the rail such that all the testing area are
0.3
covered by the mesh network. The location can be
0.2
WCL either calculated locally by the smart phone which is
PF−O
Hybrid−O
physically connect with the RFID reader, or calculated
0.1
Hybrid−L by the server on the mesh networks.
0
0 0.1 0.2 0.3 0.4 0.5
Location Error (m)
0.6 0.7 0.8
We deploy a total of 42 RFID tags along the track.
When the LRV is moving along the tagged rail, the
Fig. 14. CDF of tracking errors RFID reader keeps querying for the tags. The obser-
vation frequency of the reader f is 20Hz. We have
the speed for the hybrid method is set to vth = tested two tracks. On one track, the train is moving
2tr·f at the speed of 15km/h. On the other track, the train
Nth = 0.3m/s. In the classroom, we deploy a laptop
as the server. The laptop is connected with a TelosB is moving at 35km/h. We adjust the power level of
mote through the USB, such that the Mica2 mote on the reader to the maximum level, level 18, so that the
the RFID reader can offload computations onto the reading range, tr = 0.8m, can satisfy the condition
server through ZigBee protocol. The bandwidth of required by WCL, which is tr > 1.1a, where a is about
the ZigBee connection between the RFID reader and 0.7m.
server is measured every 5 seconds. The duration Nd (measured by the number of
To evaluate the location error, the real location time slots) that a tag is able to stay in the reading
is calculated through the integral of the preset range of the reader is respectively 3 and 7 in case
speed. The location is computed periodically us- of v = 35km/h and v = 15km/h. For one tag, if
ing the four tracking methods, WCL-Local (WCL), all the N detections fail, this tag will definitely fail
PL-Offloading (PL-O), Hybrid-Local (Hybrid-L) and to be detected; if any one of the Nd detections is
Hybrid-Offloading (Hybrid-O). For each method, we successful, this tag will be detected successfully. We
repeat the experiment 10 times by choosing different found that the detection rate of tags in the two tests
trajectories in the classroom. Fig.13 shows the 10 are quite different. The detection rate is about 70%
trajectories generated in the 4m×6m classroom. The when v = 15km/h. When v = 35km/h, the detection
result is averaged over the 10 experiments. Table 5 rate decreases to 45%.
presents the average location error and execution time In our experiment, Nth is set to an experience
per step of the three methods. Fig.14 shows the CDF of value, 30. The LVR’s speed in both the two tests are
tracking errors under the four algorithms. The hybrid above the threshold. According to the rules of the
method (Hybrid-O) almost achieve the same accuracy hybrid method, the location should be estimated by
as PF (PF-O), but outperforms PF a lot in term of particle filter. In order to evaluate the location error,
computational cost. the real location is also approximated by the integral
of the LRV’s speed. By performing the PF algorithm
based on the data set, we obtain that the location
5.2 LRV tracking in Hong Kong MTR depot error is 0.45m when v = 15km/h. However, when
In the scenario of LRV tracking, we installed an RFID v increases to 35km/h, the location error increases
reader below the LRV, shown in Fig.15. Note that to 1.2m. We also evaluate the computational cost of
our RRU1861 URF RFID reader/attenna does not the PF algorithm in this LRV tracking application. We
have computation and networking capability. To use compare the computation cost of PF in the case the
it in location estimation, we usually need to attach algorithm is performed locally on the smart phone
it with external devices. In the wheelchair tracking and the case the algorithm is offloaded onto the server
experiment, we attach it with the Mica2 mote, while in on the mesh network. We find the offloading strategy
this experiment, we attach it with one Sharp SH631W reduce the computation cost per time slot from 137
smart phone. We deploy our self-made mesh routers ms to 24.8 ms.
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
z
the targets move in a hexahedron, e.g., a warehouse,
a reading room, etc. The passive tags with known
positions are deployed on the floor and ceil. The tags’
locations are represented by (x, y, z) three coordinates.
z
©
The z coordinate of tags on the floor is zero, and
y ¶
y
the z coordinate of tags on the ceil is the height
© of the hexahedron. Fig.16 illustrates the 3-D object
tracking environment which has been discussed in
x
¶
our previous work [28]. We assume that the reading
x
range of the RFID reader is sphere. The RFID sensing
model in 3-D scenes has the same formulation with
Equation (1), where the r represents the 3-D distance
Fig. 16. RFID based object tracking in 3-D environ- between the tag and the reader. The system model
ment which represents the evolution of the target’s location
with time should change from Equation (8-10) to the
6 D ISCUSSION Equation (14-18). With the new 3-D system model,
This paper not only provides a hybrid method of the algorithms discussed in our paper for 2-D object
object tracking, but also provides a guide for users tracking can be also applied in the 3-D environment.
to deploy the RFID based tracking system under v
different performance requirements. A tradeoff be- xk = xk−1 + cos θk−1 cos φk−1 (14)
f
tween the accuracy and efficiency can be achieved by
dynamically adjusting vth or Nth in the runtime. If v
yk = yk−1 + cos θk−1 sin φk−1 (15)
the computational cost is critical in the system, a low f
value of Nth is preferred. If high accuracy is desired, v
zk = zk−1 + sin θk−1 (16)
a high value of Nth is chosen. f
RFID Sensing Model. In real world, the RFID sens-
φk = φk−1 + U (−∆φ, ∆φ) (17)
ing model will be quite different when the device is
performed in different environments. As the sensing θk = θk−1 + U (−∆θ, ∆θ) (18)
model in Section 2 is measured in the same configured
classroom, it is able to be used into the experiment of
wheelchair navigation. However, in the LRV tracking
7 R ELATED W ORKS
experiment, a more general sensing model, shown RFID Localization/Tracking. There exists a variety of
in Equation (13), is applied into the PF algorithm approaches to RFID-based localization. Many early
instead, where p0 is a constant. Only when one tag has systems use active tags, because the active tag is
been detected, the step of filtering out the invalidate like the traditional sensors, being able to transmit
particles will be executed according to the sensing signals, and thus the previous Radio Frequency (RF)
model. In other times, the prediction step based on based localization approaches, such as WLAN and
the estimated speed will be executed. WSN, can be easily applied into these RFID systems
( [7][15][19]. In these works, reference tags with known
p0 , r < tr; positions are deployed regularly on the covered area.
p(r) = (13) Lionel et al. [7] present LANDMARC that uses kNN
0, r > tr
technique for localizing unknown active tags. Huang
As shown in the LRVs tracking experiment, neither et al. [15] utilizes Bayesian inference to calculate the
the PF nor the hybrid method is suitable for tracking target position. Yamano et al. [19] use support vector
high speed moving objects cases, the tags detection machine to estimate the robot’s location. Our work in
rate is usually low. ( The detection rate is as low as this paper focuses on the object tracking using passive
45% when the LVR’s speed is v = 35km/h ). The bad RFID tags, due to its cheapness and widely usage in
accuracy in this scenario can be explained as follows: industry.
1) when the speed is too high, which also means Most related works on localization and/or objective
the value of Nd is too low, our method to estimate tracking based passive RFID tags use the geometrical
the speed will not work. Without information of the and range free methods to estimate the location of
speed, the efficiency of particle sampling decreases moving objects. Han et al. [11] present localization
seriously. 2) The invalidate particles are not filtered scheme for an indoor mobile robot using passive
out until the next tag is successfully detected. Thus, RFID tags. A triangular pattern of arranging the RFID
when the detection rate is low, these unfiltered invalid tags on the floor is adopted to reduce the estimation
particles will leads to large location error. error of the conventional square pattern. Park et al.
3-D Object Tracking Model. The proposed method [12] propose a method using read-time of IC tags to
can be easily extended into 3-D scenes. We consider reduce the localization error of an RFID navigation
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
system. However, both of these two literatures assume on the mobile device or on the cloud, such that the
a binary sensor model and do not consider detection cost is minimized. Related works on computation
failure of passive tags. partitioning considers the application modeling and
The literatures [3][6][13][14] consider a probabilistic optimization of various objectives. They differ in the
sensor model for the RFID reader, and use particle cost models, which consider either one of these factors
filter to estimate the location. They focus on the such as energy consumption of the mobile device,
tracking of robot, and assume that the robot has application delay, and data transmission amount on
abundant resources to run the costly particle filter the network, or the combination of these factors.
method. However, in this paper, we study more Most related works on computation offloading fo-
general scenario where the objects to be tracked are cus on the modeling, optimization and the design of
usually attached with cheap embedded RFID devices. runtime system to support the offloading. They do
Running PF on these devices may cause large latency, not evaluate to what extend computation offloading
and thus can not satisfy the real time property of the method can improve the efficiency of RFID tracking
tracking application. applications. Meanwhile, to maximize the profit of
Jeffrey et.al [27] evaluates the performance of parti- computation offloading, we believe that the design
cle filter algorithm running on various mobile devices, of offloading strategy should be considered together
and compares particle filter with other deterministic with the algorithm design of RFID tracking applica-
localization methods. They focus on applying particle tions.
filter in two localization systems, namely infrared
badge system from VersusTech and MIT Cricket ul-
8 C ONCLUSION
trasound system. Both the two localization systems
are using active sensors/beacons. In our work, we In this paper we have studied the problem of tracking
focus on the localization with passive sensors/ RFID mobile objects that are able to detect the presence of
tag. The environment is deployed with passive RFID passive RFID tags when traveling in their proximity.
tags, which can only receive signals passively from We proposed a hybrid method for achieving high
the objects. The communications between passive tags accuracy and efficiency in object tracking. Through
and readers is more prone to failures. Furthermore, numerical evaluations, we showed that our method
the collisions among multiple tags could increase the is more computational efficient than PF while guar-
probability of detection failures. The existing methods anteing the same accuracy with PF. We also imple-
like centroid based methods and particle filter can mented two real RFID based tracking systems using
not directly applied to solve the localization problem our proposed method in two applications, namely, the
based passive RFID. wheelchair navigation and LRVs tracking at one of
Phase value has been used to localize or track Hong Kong MTR depots. For the former application,
RFID tags to achieve high accuracy in recent works our method can achieve better performance than both
[29][30][31]. Pinit [29] compares the multipath profile PF and WCL in either efficiency or accuracy. For
(phase related information) between the target tag and the latter application, in most cases our method can
the reference tags, and estimate the targets location achieve good accuracy, but in some case the LRV
from its nearest references tags. Unlike existing RSS moves in a very high speed our method do not
based methods, Pinit can achieve high accuracy in perform well due to the low detection rate of the RFID
multipath or non-line-of-sight environments. Howev- tags.
er, it needs relatively dense deployment of reference
tags to guarantee high accuracy. Real time tracking of ACKNOWLEDGMENTS
objects may suffering from latency problem due to the
time in collection of tags multipath profiles. BagPos The research is partially supported by Hong Kong
[30] is anchor free, and does not need extra informa- RGC under GRF Grant 510412, and the National High-
tion from the reader or tags. However, it sacrifices the Technology Research and Development Program (863
area of feasible region. Tagoram [31] utilizes the phase Program) of China under Grant 2013AA01A212.
value of the backscattered signal to track the mobile
RFID tags in real time, and can achieve centimeter- R EFERENCES
level accuracy in real world deployments.
[1] R. Behnke, and D. Timmermann. AWCL: adaptive weighted
Related Works on Computation Offloading. Com- centroid localization as an efficient improvement of coarse
putation offloading has been widely used to solve grained localization. In Proc. Int. Workshop on Positioning, Navi-
the resource poverty problem of mobile devices. Nor- gation and Communication 2008, pages 243C250.
[2] J. Blumenthal, R. Grossmann, F. Golatowski, and D. Timmer-
mally to achieve a fine granularity for computation mann. Weighted Centroid Localization in Zigbee-based Sensor
offloading, a partitioning scheme has been proposed Networks. In Proc. IEEE Int. Symposium on Intelligent Signal
by researchers [24] [25] [26]. Given that the application Processing 2007, pages 1C6.
[3] D. Hahnel, W. Burgard, D. Fox, K. Fishkin, and M. Philipose.
is composed of a set of dependent tasks, computation Mapping and localization with RFID technology. In Proc. IEEE
partitioning is to decide whether each task is executed Int. Conf. on Robotics and Automation 2004, pages 1015C1021.
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/TMC.2014.2381232, IEEE Transactions on Mobile Computing
[4] D. Joho, C. Plagemann, and W. Burgard. Modeling RFID Signal [26] B. Chun, S. Ihm, P. Maniatis, M. Naik, and A. Patti.
Strength and Tag Detection for Localization and Mapping. In CloneCloud: elastic execution between mobile device and
Proc. IEEE Int. Conf. on Robotics and Automation 2009, pages 3160- cloud. In Proc. of Sixth Conf. on Computer Systems 2011, pages
3165. 301-314.
[5] A. Ghosh, and S. Das. Coverage and connectivity issues in wire- [27] J. Hightower, and G. Borriello. Particle Filters for Location
less sensor networks: A survey. Pervasive and Mobile Computing, Estimation in Ubiquitous Computing: A Case Study. In Proc.
Vol.4, Issue 3, pages 303C334, June 2008. of Int. Conf. on Ubicomp 2004.
[6] E. Giampaolo, and F. Martinelli. Robot localization by sparse [28] W. Zhu, J. Cao, Y. Xu, L. Yang, and J. Kong. Fault-Tolerant
and passive RFID tags. In Proc. IEEE Int. Symposium on Industrial RFID Reader Localization Based on Passive RFID Tags. IEEE
Electronics 2010, pages 1937C1942. Trans. on Parallel and Distributed Systems, Vol.25, Issue 8, pages
[7] L. Ni, Y. Liu, Y. Lau, and A. Patil. LANDMARC: indoor location 2065 - 2076, 2014.
sensing using active RFID. In Proc. IEEE Int. Conf. on Pervasive [29] J. Wang and D. Katabi. Dude, Wheres My Card? RFID Po-
Computing and Communications 2003, pages 407C415. sitioning that Works with Multipath and Non-line-of-sight. in
[8] M. Arulampalam, S. Maskell, N. Gordon, and T. Clapp. A Proc. of ACM SIGCOMM, 2013.
tutorial on particle filters for online nonlinear/non-gaussian [30] T. Liu, L. Yang, Q. Lin, Y. Guo, Y. Liu. Anchor-free Backscatter
bayesian tracking. IEEE Tran. on Signal Processing, Vol.50, Issue Positioning for RFID Tags with High Accuracy. in Proc. of IEEE
2, pages 174C188, 2002. INFOCOM, 2014.
[9] N. Bulusu, J. Heidemann, and D. Estrin. Gps-less low cost [31] L. Yang, Y. Chen, X. Li, C. Xiao, M. Li, Y. Liu. Tagoram: Real-
outdoor localization for very small devices. IEEE Personal Com- Time Tracking of Mobile RFID Tags to High Precision Using
munications Magazine, Vol.7, Issue 5, pages 28C34, 2000. COTS Devices. in Proc. of ACM MobiCom, 2014.
[10] N. Bulusu, J. Heidemann, and D. Estrin. Centroid localization
of uncooperative nodes in wireless networks using a relative
span weighting method. EURASIP Journal on Wireless Commu-
nications and Networking, Vol.10, Issue 11, pages 1C11, 2010.
[11] S. Han, H. Lim and J. Lee. An efficient localization scheme
for a differential-drive mobile robot based on rfid system. IEEE Lei Yang received the BSc degree in elec-
Trans. On Industrial Electronics, Vol.54, Issue 6, pages 362C369, tronic engineering from Wuhan University,
2007. P.R.China, in 2007, and the MSc degree in
[12] S. Park and S. Hashimoto. Autonomous mobile robot naviga- computer science from Institute of Comput-
tion using passive rfid in indoor environment. IEEE Trans. on ing Technology, Chinese Academy of Sci-
Industrial Electronics, Vol.56, Issue 7, pages 2366C2373, 2009. ence, P.R.China, in 2010. He is a post-
[13] S. Schneegans, and P. Vorst. Using RFID snapshots for mobile doctoral fellow of the Department of Comput-
robot self-localization. In Proc. European Conference on Mobile ing at Hong Kong Polytechnic University. His
Robots 2007, pages 1C6. research interest includes mobile cloud com-
[14] P. Vorst, and A. Zell. Semi-autonomous learning of an RFID puting, RFID systems, and social computing.
sensor model for mobile robot self-localization. In Proc. Euro-
pean Robotics Symposium 2008, pages 273C282.
[15] X. Huang, R. Janaswamy, and A. Ganz. Scout: Outdoor lo-
calization using active RFID technology. In Proc. Int. Conf. on Jiannong Cao received the BSc degree from
Broadband Communications, Networks and Systems 2006, pages Nanjing University, China, in 1982, and the
1C10. MSc and PhD degrees from Washington S-
[16] X. Liu, M. Corner, and P. Shenoy. Ferret: RFID localization for tate University, USA, in 1986 and 1990, all
pervasive multimedia. In Proc. ACM Int. Conf. on Ubiquitious in computer science. He is currently a chair
Computing 2006, pages 422C440. professor and the head of the Department
[17] Y. Zou and K. Chakrabarty. Sensor deployment and target of Computing at Hong Kong Polytechnic Uni-
localization in distributed sensor networks. IEEE Trans. on versity. His research interests include paral-
Embedded Computing and System, Vol.3, Issue 1, pages 61C91, lel and distributed computing, computer net-
2004. works, mobile and pervasive computing, fault
[18] K. Yamano, K. Tanaka, M. Hirayama, and et al. Self- tolerance, and middle-ware.
localization of mobile robots with RFID system by using sup-
port vector machine. In Proc. IEEE/RSJ Int. Conf. on Intelligent
Robots and Systems 2004, pages 3756C3761. Weiping Zhu received the BSc degree from
[19] M. Richard, and S. Nozer. Understanding the Kalman Filtr. The Chongqing University, China, in 2003, the
American Statistician, Vol.37, Issue 2, 1983 MSc degree from Shanghai JiaoTong Uni-
[20] J. Cao, K. Xie, W. Wu, C. Liu, and et al. HAWK: Real-world versity, China, in 2006, and the PhD de-
Implementation of High-performance Heterogeneous Wireless gree from Hong Kong Polytechnic University,
Network for Internet Access. In Proc. of Int. Workshop on Next Hong Kong, in 2013, all in computer science.
Generation Network Architectures 2009, pages 214-220. He is an associate professor in the Interna-
[21] Z. Li, C. Wang, and R. Xu. Computation offloading to save tional School of Software at Wuhan Universi-
energy on handheld devices: a partition scheme. In Proc. of ty. His research interest includes RFID, wire-
Int. Conf. on Compilers, Architecture, and Synthesis for Embedded less sensor networks, distributed computing,
Systems 2011, pages 238-246. and pervasive computing.
[22] K. Kumar, and Y. Lu. Cloud computing for mobile users:
Can offloading computation save energy. IEEE Computer, Vol.43,
Issue 4, pages 51-56, 2009. Shao-Jie Tang received his Ph.D degree
[23] M. Satyanarayanan, and P. Bahl, R. Caceres, and N. Davies. from Department of Computer Science at
Cloud computing for mobile users: Can offloading computation Illinois Institute of Technology in 2012. He re-
save energy. IEEE Pervasive Computing, Vol.6, Issue 4, pages 12- ceived B.S. in Radio Engineering from South-
23, 2009. east University, P.R. China in 2006. He is
[24] L. Yang, J. Cao, S. Tang, T. Li, and A. Chan. A framework an assistant professor with Naveen Jindal
for partitioning and execution of data stream applications in School of Management, University of Texas
mobile cloud computing. In Proc. of IEEE Int. Conf. on Cloud at Dallas. His main research interests focus
Computing 2012, pages 794-782. on wireless networks (including sensor net-
[25] E. Cuervoy, A. Balasubramanianz, and D. Cho. MAUI: making works and cognitive radio networks), social
smartphones last longer with code offload. In Proc. of ACM Int. networks, pervasive computing, mobile cloud
Conf. on Mobile Systems, Applicaitons and Services 2010, pages computing and algorithm analysis and design.
49-62.
(c) 2015 Crown Copyright. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.